Take-home Exercise 2

0. Goals & Objectives.

Our goal is to conduct a case study to demonstrate the potential value of Geospatial Data Science & Analysis, by integrating publicly available data from multiple sources to build a spatial interaction models, to determine factors affecting urban mobility patterns of public bus transit.

Recentdeployments of massively pervasive computing technologies such as in-vehicle GPS and SMART cards by public transport provide plenty of data for tracking movement patterns through space and time, but the explosive growth of data has outstripped public services’ ability to utilise, transform, and understand the data.

More detail about the task from: https://isss624-ay2023-24nov.netlify.app/take-home_ex02

Modifible Areal Unit Problem (MAUP): - MPSZ is too coarse, too huge a subzone area; people may live only in a corner of the subzone - Planning subzones too large, so we use analytical hexagon

“Map is not interesting, pattern revealed by and factors affecting the map is interesting. Can we explain this by building a spatial model?” - Building a model to explain flows; Spatial Interaction Model

GLM: Generalised Linear-Regression Model (over linear model):

  • Dwelling Units as proxy for population;
  • Can compare HDB-only VS dwelling units vs room-flat vs 1/2/3/ room flat unit
  • POI: points of interest name & type
Important Note

Due to the nature of EDA and Data Analysis, parts of this page have been Collapsed or placed behind tabs, to avoid excessive scrolling.

For easier reading, this study is also presented in point-form.

0.1 Dataset used

0.1.1 Aspatial Dataset

Dataset, Purpose & Source: Key columns

hdb.csv : HDB Property information data with geocoding

Via data.gov
  • lat, long: coordinates |
  • max_floors, total_dwelling_units : idea of how many units each HDB holds

origin_destination_bus_202310 : Volume of bus passenger trips, by origin/destination, data prepared for 2023 October

Via LTA DataMall
  • ORIGIN_PT_code, DESTINATION_PT_code: O/D information on bus trip
  • TIME_PER_HOUR, DAY_TYPE: Time/Date data on when the trips were taken
  • TOTAL_TRIPS: Bus trip passenger volume

0.1.2 Geospatial Dataset

Sample grid table. |
Filename, Purpose & Source: Key columns

hdb.csv : HDB Property information data with geocoding

Via data.gov
  • lat, long: coordinates |
  • max_floors, total_dwelling_units : idea of how many units each HDB holds

origin_destination_bus_202310 : Volume of bus passenger trips, by origin/destination, data prepared for 2023 October

Via LTA DataMall
  • ORIGIN_PT_code, DESTINATION_PT_code: O/D information on bus trip
  • TIME_PER_HOUR, DAY_TYPE: Time/Date data on when the trips were taken
  • TOTAL_TRIPS: Bus trip passenger volume

1. Geospatial Data Wrangling

This study was performed in R, written in R Studio, and published using Quarto.

1.1 Import Packages

This function calls pacman to load sf, tidyverse, tmap, knitr packages;

  • tmap : For thematic mapping; powerful mapping package;
  • sf : for geospatial data handling, but also geoprocessing: buffer, point-in-polygon count, etc;
  • sfdep : useful functions for creating weight matrix, LISA calculations etc;
  • tidyverse : for non-spatial data handling; commonly used R package and contains dplyr for dataframe manipulation and ggplot for data visualization;
  • DT: for displaying datatables;
  • leaflet: for custom layer controls over tmap visualisations.
  • stplanar: for creating desire lines to visualize O/D flows
pacman::p_load(tmap, sf, sfdep, tidyverse, DT, leaflet, stplanr )

1.2 Import Geospatial Data

1.2.1 raw_bus_stop_sf: Load Geospatial Bus Stop Data

  • First, we load BusStop shapefile data from LTA Datamall;
  • st_read() is used to import the ESRI Shapefile data into an sf dataframe.
    • From previous take-home exercise, we know BusStop has the incorrect CRS (coordinate reference system), as EPSG 9001 instead of 3414, so we use st_set_crs() to correct this
    • We use head() to preview the first 6 rows
show code 
raw_bus_stop_sf <- st_read(dsn = "data/geospatial", 
                 layer = "BusStop") %>%
  st_set_crs(3414)
Reading layer `BusStop' from data source 
  `C:\1darren\ISSS624\Take-home_Ex02\data\geospatial' using driver `ESRI Shapefile'
Simple feature collection with 5161 features and 3 fields
Geometry type: POINT
Dimension:     XY
Bounding box:  xmin: 3970.122 ymin: 26482.1 xmax: 48284.56 ymax: 52983.82
Projected CRS: SVY21
Warning: st_crs<- : replacing crs does not reproject data; use st_transform for
that
show code 
head(raw_bus_stop_sf)
Simple feature collection with 6 features and 3 fields
Geometry type: POINT
Dimension:     XY
Bounding box:  xmin: 13228.59 ymin: 30391.85 xmax: 41603.76 ymax: 44206.38
Projected CRS: SVY21 / Singapore TM
  BUS_STOP_N BUS_ROOF_N             LOC_DESC                  geometry
1      22069        B06   OPP CEVA LOGISTICS POINT (13576.31 32883.65)
2      32071        B23         AFT TRACK 13 POINT (13228.59 44206.38)
3      44331        B01              BLK 239  POINT (21045.1 40242.08)
4      96081        B05 GRACE INDEPENDENT CH POINT (41603.76 35413.11)
5      11561        B05              BLK 166 POINT (24568.74 30391.85)
6      66191        B03         AFT CORFE PL POINT (30951.58 38079.61)
  • We check the coordinate reference system with st_crs(); and see that it is now indeed correctly set to 3414:
show code 
st_crs(raw_bus_stop_sf)
Coordinate Reference System:
  User input: EPSG:3414 
  wkt:
PROJCRS["SVY21 / Singapore TM",
    BASEGEOGCRS["SVY21",
        DATUM["SVY21",
            ELLIPSOID["WGS 84",6378137,298.257223563,
                LENGTHUNIT["metre",1]]],
        PRIMEM["Greenwich",0,
            ANGLEUNIT["degree",0.0174532925199433]],
        ID["EPSG",4757]],
    CONVERSION["Singapore Transverse Mercator",
        METHOD["Transverse Mercator",
            ID["EPSG",9807]],
        PARAMETER["Latitude of natural origin",1.36666666666667,
            ANGLEUNIT["degree",0.0174532925199433],
            ID["EPSG",8801]],
        PARAMETER["Longitude of natural origin",103.833333333333,
            ANGLEUNIT["degree",0.0174532925199433],
            ID["EPSG",8802]],
        PARAMETER["Scale factor at natural origin",1,
            SCALEUNIT["unity",1],
            ID["EPSG",8805]],
        PARAMETER["False easting",28001.642,
            LENGTHUNIT["metre",1],
            ID["EPSG",8806]],
        PARAMETER["False northing",38744.572,
            LENGTHUNIT["metre",1],
            ID["EPSG",8807]]],
    CS[Cartesian,2],
        AXIS["northing (N)",north,
            ORDER[1],
            LENGTHUNIT["metre",1]],
        AXIS["easting (E)",east,
            ORDER[2],
            LENGTHUNIT["metre",1]],
    USAGE[
        SCOPE["Cadastre, engineering survey, topographic mapping."],
        AREA["Singapore - onshore and offshore."],
        BBOX[1.13,103.59,1.47,104.07]],
    ID["EPSG",3414]]
  • We use qtm() to perform a quick visualization of the various bus stops:
show code 
tmap_mode("plot")
tmap mode set to plotting
show code 
qtm(raw_bus_stop_sf)

1.2.2 mpsz_sf: Visualizing Singapore’s Master Plan 2019 Subzone Boundaries

  • We now load Master Plan 2019 Subzone Boundary

  • Next, we load MPSZ-2019 shapefile data from Data.gov.sg, comprising Master Plan 2019 Subzone Boundary (No Sea) data ;

  • st_read() is used to import the ESRI Shapefile data into an sf dataframe;

    • We use st_crs() to check the Coordinate Reference System (CRS);
show code 
mpsz_sf <- st_read(dsn = "data/geospatial", 
                 layer = "MPSZ-2019") 
Reading layer `MPSZ-2019' from data source 
  `C:\1darren\ISSS624\Take-home_Ex02\data\geospatial' using driver `ESRI Shapefile'
Simple feature collection with 332 features and 6 fields
Geometry type: MULTIPOLYGON
Dimension:     XY
Bounding box:  xmin: 103.6057 ymin: 1.158699 xmax: 104.0885 ymax: 1.470775
Geodetic CRS:  WGS 84
show code 
st_crs(mpsz_sf)
Coordinate Reference System:
  User input: WGS 84 
  wkt:
GEOGCRS["WGS 84",
    DATUM["World Geodetic System 1984",
        ELLIPSOID["WGS 84",6378137,298.257223563,
            LENGTHUNIT["metre",1]]],
    PRIMEM["Greenwich",0,
        ANGLEUNIT["degree",0.0174532925199433]],
    CS[ellipsoidal,2],
        AXIS["latitude",north,
            ORDER[1],
            ANGLEUNIT["degree",0.0174532925199433]],
        AXIS["longitude",east,
            ORDER[2],
            ANGLEUNIT["degree",0.0174532925199433]],
    ID["EPSG",4326]]
  • We see that MPSZ-2019 is set to WGS 84/EPSG 4326 projection, which is inconsistent with the rest of our dataset;
  • Thus, we use st_transform() to reproject the data to SVY21/EPSG 3414 Coordinate Reference System (CRS) to ensure consistent distance calculation:
show code 
mpsz_sf <- mpsz_sf %>%
  st_transform(crs=3414)
st_crs(mpsz_sf)
Coordinate Reference System:
  User input: EPSG:3414 
  wkt:
PROJCRS["SVY21 / Singapore TM",
    BASEGEOGCRS["SVY21",
        DATUM["SVY21",
            ELLIPSOID["WGS 84",6378137,298.257223563,
                LENGTHUNIT["metre",1]]],
        PRIMEM["Greenwich",0,
            ANGLEUNIT["degree",0.0174532925199433]],
        ID["EPSG",4757]],
    CONVERSION["Singapore Transverse Mercator",
        METHOD["Transverse Mercator",
            ID["EPSG",9807]],
        PARAMETER["Latitude of natural origin",1.36666666666667,
            ANGLEUNIT["degree",0.0174532925199433],
            ID["EPSG",8801]],
        PARAMETER["Longitude of natural origin",103.833333333333,
            ANGLEUNIT["degree",0.0174532925199433],
            ID["EPSG",8802]],
        PARAMETER["Scale factor at natural origin",1,
            SCALEUNIT["unity",1],
            ID["EPSG",8805]],
        PARAMETER["False easting",28001.642,
            LENGTHUNIT["metre",1],
            ID["EPSG",8806]],
        PARAMETER["False northing",38744.572,
            LENGTHUNIT["metre",1],
            ID["EPSG",8807]]],
    CS[Cartesian,2],
        AXIS["northing (N)",north,
            ORDER[1],
            LENGTHUNIT["metre",1]],
        AXIS["easting (E)",east,
            ORDER[2],
            LENGTHUNIT["metre",1]],
    USAGE[
        SCOPE["Cadastre, engineering survey, topographic mapping."],
        AREA["Singapore - onshore and offshore."],
        BBOX[1.13,103.59,1.47,104.07]],
    ID["EPSG",3414]]
  • Now, let’s visualize the bus stop within the mpsz subzone boundaries;
  • We set tmap_mode("plot") to allow us to scroll;
  • We use tm_shape() + tm_polygons() to map a base layer of the mpsz boundaries;
    • On top of which, we layer tm_shape() + tm_dots() to show the bus stops.
show code 
tmap_mode("plot")
tmap mode set to plotting
show code 
tmap_options(check.and.fix = TRUE)
tm_shape(mpsz_sf) +
  tm_polygons() + 
  tm_shape(raw_bus_stop_sf) + 
  tm_dots()
Warning: The shape mpsz_sf is invalid. See sf::st_is_valid

  • We note there are a number of bus stops outside Singapore’s boundaries; Specifically, three bus stops in a cluster just outside the Causeway, and one further North.
  • We perform several steps to isolate & check the data;
    • we use st_filter() to find bus stops within Singapore’s Administrative National Boundaries, and create sg_bus_stop_sf for future use.
    • to check what bus stops have been dropped, we use anti_join() to compare raw_bus_stop_sf with sg_bus_stop_sf. We use st_drop_geometry on both sf dataframes to only compare the non-geometry columns.
show code 
sg_bus_stop_sf <- st_filter(raw_bus_stop_sf, mpsz_sf)
anti_join(st_drop_geometry(raw_bus_stop_sf), st_drop_geometry(sg_bus_stop_sf))
Joining with `by = join_by(BUS_STOP_N, BUS_ROOF_N, LOC_DESC)`
  BUS_STOP_N BUS_ROOF_N            LOC_DESC
1      47701        NIL          JB SENTRAL
2      46239         NA          LARKIN TER
3      46609         NA     KOTARAYA II TER
4      46211        NIL JOHOR BAHRU CHECKPT
5      46219        NIL JOHOR BAHRU CHECKPT
  • We see there are in fact 5 bus stops outside of Singapore (including the defunct Kotaraya II Terminal) that have been removed, which means our data cleaning was correct.

1.3 Geospatial Data Cleaning

1.3.1 Removing Duplicate Bus Stops

  • From Take-home Exercise 1, we know that there are a number of repeated bus stops. We repeat some steps;
  • We use length() to find the total number of raw values in the $BUS_STOP_N column of sg_bus_stop_sf;
    • We then compare this to length(unique()) to find the unique values;
  • And, indeed, we find there are 16 bus stops that are repeated;
cat("\nResults before removing duplicates: \n=======================================================\n")

Results before removing duplicates: 
=======================================================
cat("Total number of rows in sg_bus_stop_sf\t\t: ", paste0(length(sg_bus_stop_sf$BUS_STOP_N)))
Total number of rows in sg_bus_stop_sf      :  5156
cat("\nTotal unique bus stop IDs in sg_bus_stop_sf\t: ", paste0(length(unique(sg_bus_stop_sf$BUS_STOP_N))))

Total unique bus stop IDs in sg_bus_stop_sf :  5140
cat("\nRepeated bus stops\t\t\t\t:   ", paste0(length(sg_bus_stop_sf$BUS_STOP_N) - length(unique(sg_bus_stop_sf$BUS_STOP_N))))

Repeated bus stops              :    16
  • It appears there are 16 datapoints that are specifically repeated; let’s remove them by deleting the duplicated rows:
    • we use duplicated() to identify rows with repeated values of $BUS_STOP_N; duplicated rows will return TRUE while all other rows will return FALSE
    • We use ! to invert the values, so only the unduplicated rows will return TRUE.
    • We then use square brackets [] to index sg_bus_stop_sf based on the rows, and return only the unduplicated rows;
    • We then assign the output using <- into bus_stop_sf, for use.
show code 
bus_stop_sf <- sg_bus_stop_sf[!duplicated(sg_bus_stop_sf$BUS_STOP_N), ]
head(bus_stop_sf)
Simple feature collection with 6 features and 3 fields
Geometry type: POINT
Dimension:     XY
Bounding box:  xmin: 13228.59 ymin: 30391.85 xmax: 41603.76 ymax: 44206.38
Projected CRS: SVY21 / Singapore TM
  BUS_STOP_N BUS_ROOF_N             LOC_DESC                  geometry
1      22069        B06   OPP CEVA LOGISTICS POINT (13576.31 32883.65)
2      32071        B23         AFT TRACK 13 POINT (13228.59 44206.38)
3      44331        B01              BLK 239  POINT (21045.1 40242.08)
4      96081        B05 GRACE INDEPENDENT CH POINT (41603.76 35413.11)
5      11561        B05              BLK 166 POINT (24568.74 30391.85)
6      66191        B03         AFT CORFE PL POINT (30951.58 38079.61)
show code 
cat("\nResults after removing duplicates: \n=======================================================\n")

Results after removing duplicates: 
=======================================================
show code 
cat("Total number of rows in bus_stop_sf\t\t: ", paste0(length(bus_stop_sf$BUS_STOP_N)))
Total number of rows in bus_stop_sf     :  5140
show code 
cat("\nTotal unique bus stop IDs in bus_stop_sf\t: ", paste0(length(unique(bus_stop_sf$BUS_STOP_N))))

Total unique bus stop IDs in bus_stop_sf    :  5140
show code 
cat("\nRepeated bus stops\t\t\t\t:   ", paste0(length(bus_stop_sf$BUS_STOP_N) - length(unique(bus_stop_sf$BUS_STOP_N))))

Repeated bus stops              :    0
  • We can do a quick check to visualize these;
    • We use tmap_mode("view") to allow us to scroll around and check that the bus stops fall within Singapore’s national boundaries, and set zoom limits to focus the attention
    • We combine tm_shape() and tm_polygons() to map the master plan subzones in grey;
    • We combine tm_shape() and tm_dots() to map locations of bus stops; for visual distinction with the grey zones, we use the “Spectral” palette
show code 
tmap_mode("view")
tmap mode set to interactive viewing
show code 
tm_shape(mpsz_sf) +
  tm_polygons() + 
  tm_shape(bus_stop_sf) + 
  tm_dots(col = "BUS_STOP_N", palette = "Spectral", legend.show = FALSE) + 
  tm_view(set.zoom.limits = c(11, 13))
Warning: The shape mpsz_sf is invalid (after reprojection). See sf::st_is_valid
Warning: Number of levels of the variable "BUS_STOP_N" is 5140, which is larger
than max.categories (which is 30), so levels are combined. Set
tmap_options(max.categories = 5140) in the layer function to show all levels.
show code 
tmap_mode("plot")
tmap mode set to plotting
  • Now, we can start preparing the hexagon map.

1.4 Generating Hexagon Maps

  • We generate the hexagon map in three steps:
  1. We use st_make_grid() with square = FALSE to create the hexagon layer as defined in the study, which we name raw_hex_grid;
    • We pass cellsize = 750 to create the hexagons of necessary size. Prof Kam defined the apothem as 375m, as the Traffic Analysis Zone is typically 750m in size.
    • I used units::as_units to pass 750 metres into the argument. I am still uncertain whether a length of 750m needs to be reprojected, or whether we need to do any further transformation.
    • We use st_transform() just in case we need to reproject the coordinate system, just in case.
  2. We use st_sf() to convert raw_hex_grid into an sf dataframe for further manipulation later; - However, trying to visualize this right now just gives us a map full of hexagons, so need to eliminate the empty hexagons; - We use mutate() to create a new column, $n_bus_stops that counts the number of bus stops in each hexagon using lengths(st_intersects())
    • st_intersects() gives us a list of bus stops in each hexagon, so we use lengths() to count the number
  3. We create our final sf dataframe, hexagon_sf in two steps; - First, we use filter() to select only hexagons with nonzero number of bus stops; - Then, mutate() is used here to create a grid_id column, labelling only the hexagons with nonzero bus stops.
  • Finally, we perform a quick plot to confirm that every bus stop is inside a hexagon;
    • Using hexagon_sf as our base layer, we use tmap_shape() and tm_polygons() to visualizes our hexes. We pass palette = "Spectral" for visual distinguishment against the black dots of the bus stops.
    • We use tmap_shape() and tm_dots() to visualize our bus stop as black dots
show code 
# STEP 1 - Create hexagon map 
raw_hex_grid <- st_make_grid(bus_stop_sf, cellsize = units::as_units(375, "m"), what = "polygons", square = FALSE) %>%
  st_transform(crs = 3414)

# STEP 2 - Convert to sf object and count the number of bus stops inside;
raw_hex_grid <- st_sf(raw_hex_grid) %>%
  mutate(n_bus_stops = lengths(st_intersects(raw_hex_grid, sg_bus_stop_sf))) 

# Count number of points in each grid, code snippet referenced from: 
# https://gis.stackexchange.com/questions/323698/counting-points-in-polygons-with-sf-package-of-r


# STEP 3 - Filter for nonempty hexes and label:

hexagon_sf <- filter(raw_hex_grid, n_bus_stops > 0) 
hexagon_sf <- hexagon_sf %>%
  mutate(grid_id = 1:nrow(hexagon_sf)) %>%
  select(grid_id, raw_hex_grid, n_bus_stops)

tmap_mode("plot")
tmap mode set to plotting
show code 
tm_shape(hexagon_sf) +
  tm_polygons(col = "grid_id", palette = "Spectral", legend.show = FALSE) + 
  tm_shape(bus_stop_sf) + 
  tm_dots()

show code 
tmap_mode("plot")
tmap mode set to plotting
  • We can visually confirm that every black dot is within a hexagon.
  • We re-use a bit of code from Take-Home Exercise 1 to plot our analytical hexagons on a map of Singapore and visualize the number of bus stops in each hexagon:
show code 
tmap_mode("view")
tmap mode set to interactive viewing
show code 
tm_basemap(providers$OneMapSG.Grey) + 
  tm_shape(hexagon_sf) +
  tm_fill(
    col = "n_bus_stops",
    palette = "-plasma",
    style = "cont",
    
    breaks = c(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12),
    title = "Number of bus_stops",
    id = "grid_id",
    showNA = FALSE,
    alpha = 0.6,
    popup.vars = c(
      "Number of Bus Stops: " = "n_bus_stops"
    ),
    popup.format = list(
      n_bus_stops = list(format = "f", digits = 0)
    )
  ) +
  tm_borders(col = "grey40", lwd = 0.7)  + 
  tm_view(set.zoom.limits = c(11, 13))
show code 
tmap_mode("plot")
tmap mode set to plotting
  • We perform some simple stats to count the total number of filtered hexagons, and to see the maximum number of bus stops in a hexagon.
    • Unlike Take-Home Exercise 1, the number of hexagons have decreased, and the maximum number of bus stops per hexagon is lower.
show code 
# NB: Code reused from my own take-home exercise 1
cat(paste("Total number of raw hexagons is\t\t\t: ", nrow(raw_hex_grid), "\n"))
Total number of raw hexagons is         :  8843
show code 
cat(paste("Total number of hexagons (n_bus_stop > 1) is\t: ", nrow(hexagon_sf)), "\n")
Total number of hexagons (n_bus_stop > 1) is    :  2172
show code 
cat("\nPrinting map_hexagon_sf:\n >> ")

Printing map_hexagon_sf:
 >>
show code 
hexagon_sf[hexagon_sf$n_bus_stops > 5, ]
Simple feature collection with 47 features and 2 fields
Geometry type: POLYGON
Dimension:     XY
Bounding box:  xmin: 10532.62 ymin: 29837.95 xmax: 42782.62 ymax: 47807.98
Projected CRS: SVY21 / Singapore TM
First 10 features:
    grid_id n_bus_stops                   raw_hex_grid
149     149           6 POLYGON ((10720.12 34059.82...
170     170           6 POLYGON ((11470.12 34059.82...
191     191           6 POLYGON ((12032.62 35683.62...
214     214           6 POLYGON ((12595.12 36008.38...
217     217           6 POLYGON ((12595.12 39255.98...
297     297           6 POLYGON ((14282.62 35034.1,...
347     347           7 POLYGON ((15782.62 34384.58...
405     405           6 POLYGON ((17470.12 39905.49...
426     426           6 POLYGON ((17845.12 39905.49...
441     441           6 POLYGON ((18032.62 40230.25...
  • For the next step, we’ll need to manage the aspatial bus trips dataset, which is what we’ll work on now.

1.5 Aspatial Data Wrangling: Bus trip dataset

1.5.1 Import Bus O/D Dataset

  • For our purposes, we will focus only on 2023-October Passenger Volume by Origin Destination Bus Stops, downloaded from LTA DataMall;

  • We use read_csv() to load the data from the .csv file;

  • We use select() with a - sign to remove two columns redundant for our analysis:

    • $PT_TYPE column indicates the type of public transport, however, every value is “BUS”
    • $YEAR_MONTH column similarly has “2023-10” for every value, which we are aware of
    • With this in mind, we drop these two columns to save space.

  • Finally, we use mutate_at() to convert two columns ($ORIGIN_PT_CODE and $DESTINATION_PT_CODE)from character to factor, for easier analysis.

  • We use str() to check the columns, datatypes, and number of rows:

show code 
odbus_2310 <- read_csv("data/aspatial/origin_destination_bus_202310.csv") %>%
  select( -PT_TYPE, -YEAR_MONTH) %>%
  mutate_at(c("ORIGIN_PT_CODE", "DESTINATION_PT_CODE"), as.factor)
Rows: 5694297 Columns: 7
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr (5): YEAR_MONTH, DAY_TYPE, PT_TYPE, ORIGIN_PT_CODE, DESTINATION_PT_CODE
dbl (2): TIME_PER_HOUR, TOTAL_TRIPS

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
show code 
str(odbus_2310)
tibble [5,694,297 × 5] (S3: tbl_df/tbl/data.frame)
 $ DAY_TYPE           : chr [1:5694297] "WEEKENDS/HOLIDAY" "WEEKDAY" "WEEKENDS/HOLIDAY" "WEEKDAY" ...
 $ TIME_PER_HOUR      : num [1:5694297] 16 16 14 14 17 17 17 7 14 14 ...
 $ ORIGIN_PT_CODE     : Factor w/ 5073 levels "01012","01013",..: 105 105 4428 4428 2011 834 834 781 4462 4462 ...
 $ DESTINATION_PT_CODE: Factor w/ 5077 levels "01012","01013",..: 240 240 4742 4742 693 809 809 235 4002 4002 ...
 $ TOTAL_TRIPS        : num [1:5694297] 3 5 3 5 4 1 24 2 1 7 ...
  • This is a huge tibble dataframe with over 5 million rows, so we will filter this now by peaks;
  • For this study, we focus on Weekday afternoon peak

1.5.2 Filter For Peaks – Weekday Afternoon

  • We now perform a multi-step filter process;
    1. We combine mutate() with case_when() to create a new column, $PEAK, based on the study criteria:
      1. We set the value to “WEEKDAY_AFTERNOON_PEAK” if $DAY_TYPE is “WEEKDAY” and bus tap-on time (e.g. $TIME_PER_HOUR) is between 5 pm and 8pm, inclusive;
        • Note that we convert the values for $TIME_PER_HOUR to 24-hour clock, e.g. “5pm” is “17” hundred hours, “8pm” is “20” hundred hours.
      2. For all remaining values, we assign an “Unknown” value.
    2. We then use filter() to eliminate those with “Unknown” $PEAK values, i.e. rows outside the period of interest for the study
    3. We use group_by() to group the values by $ORIGIN_PT_CODE and $DESTINATION_PT_CODE, and use summarise() to aggregate the sum of $TOTAL_TRIPS as a new column, $TRIPS.
    4. We use write_rds() to save the output dataframe, odbus_filtered, as an RDS object for future reference/load.
odbus_filtered <- odbus_2310 %>%
  mutate(PEAK = case_when(
    DAY_TYPE == "WEEKDAY" & TIME_PER_HOUR >= 17 &  TIME_PER_HOUR <= 20 ~ "WEEKDAY_AFTERNOON_PEAK",
    TRUE ~ "Unknown"
  )) %>%
  filter(
    case_when(
      PEAK == "Unknown" ~ FALSE,
      TRUE ~ TRUE
    )) %>%
  group_by(ORIGIN_PT_CODE, DESTINATION_PT_CODE) %>%
  summarise(TRIPS = sum(TOTAL_TRIPS))
`summarise()` has grouped output by 'ORIGIN_PT_CODE'. You can override using
the `.groups` argument.
write_rds(odbus_filtered, "data/rds/odbus_filtered_weekday.rds")
head(odbus_filtered)
# A tibble: 6 × 3
# Groups:   ORIGIN_PT_CODE [1]
  ORIGIN_PT_CODE DESTINATION_PT_CODE TRIPS
  <fct>          <fct>               <dbl>
1 01012          01112                 540
2 01012          01113                 516
3 01012          01121                 303
4 01012          01211                 363
5 01012          01311                 630
6 01012          01549                   6

1.6 Combine Bus Trip Data With hexagon_sf Dataframe

  • For our study purposes, we need to have the number of bus trips originating from each hexagon. In order to achieve this, we must combine our three current dataframes:
    • hexagon_sf, an sf dataframe with $grid_id column (primary key) and the specific polygon geometry of each hexagon;
    • bus_stop_sf, an sf dataframe with the $BUS_STOP_N (primary key) and the point geometry of each bus stop;
    • odbus_filtered, a tibble dataframe with the $ORIGIN_PT_CODE (primary key) column and the trip details for each of the four peak periods of interest.

1.6.1 bus_stop_hexgrid_id: Identify Hexagon grid_id For Each Bus Stop

  • First, we combine hexagon_sf with sg_bus_stop_sf ;
    • We use st_intersection to combine the dataframes such that each row of sg_bus_stop_sf has an associated grid_id;
    • We use select() to filter the resultant bus_stop_hexgrid_id dataframe to only $grid_id and $BUS_STOP_N columns, and use st_drop_geometry() to convert into a simple dataframe with just two columns:
show code 
bus_stop_hexgrid_id <- st_intersection(bus_stop_sf, hexagon_sf) %>%
  select(BUS_STOP_N, grid_id) %>%
  st_drop_geometry()
Warning: attribute variables are assumed to be spatially constant throughout
all geometries
show code 
cat("\nNumber of bus stops\t:", length(unique(bus_stop_hexgrid_id$BUS_STOP_N)))

Number of bus stops : 5140
show code 
cat("\nNumber of hexgrids\t:", length(unique(bus_stop_hexgrid_id$grid_id)))

Number of hexgrids  : 2170
show code 
head(bus_stop_hexgrid_id)
       BUS_STOP_N grid_id
3265        25059       1
3265.1      25059       2
2566        25751       3
254         26379       4
2399        26369       5
2893        25761       6
  • Eagle-eyed readers may notice that BUS_STOP_N #25059 is in both $grid_id 1 & 2;
  • Let’s check if there’s any other duplicates:
show code 
bus_stop_hexgrid_id$BUS_STOP_N[duplicated(bus_stop_hexgrid_id$BUS_STOP_N)]
[1] "25059"
  • Only BUS_STOP_N #25059 is duplicated; let’s see why this is:
show code 
tmap_mode("view")
tmap mode set to interactive viewing
show code 
tm_shape(hexagon_sf[hexagon_sf$grid_id %in% c(1, 2),]) +
  tm_polygons(col = "grid_id", palette = "Spectral") + 
  tm_shape(bus_stop_sf[bus_stop_sf$BUS_STOP_N %in% c(25059),]) + 
  tm_dots()
show code 
tmap_mode("plot")
tmap mode set to plotting
  • It appears that BUS_STOP_N #25059 is exactly on the boundary between the two hexagons.
  • Here we perform simple manual intervention data cleaning, declaring the bus stop is in $grid_id 2 by deleting the first row;
    • We use subset() to select values of $grid_id that are not equal to 1
show code 
bus_stop_hexgrid_id <- subset(bus_stop_hexgrid_id, grid_id != 1)

cat("\nNumber of bus stops\t:", length(unique(bus_stop_hexgrid_id$BUS_STOP_N)))

Number of bus stops : 5140
show code 
cat("\nNumber of hexgrids\t:", length(unique(bus_stop_hexgrid_id$grid_id)))

Number of hexgrids  : 2169
show code 
cat("\n\nPrinting rows with duplicated bus stop values\t:", bus_stop_hexgrid_id$BUS_STOP_N[duplicated(bus_stop_hexgrid_id$BUS_STOP_N)])


Printing rows with duplicated bus stop values   :
show code 
cat("\n\n(If empty, it worked!)")


(If empty, it worked!)
  • Now we append hexagon code onto odbus_filtered. We do this in two steps;
  • First, we use left_join() to add the hexagon $grid_id by BUS_STOP_N number;
    • We use rename() to rename columns, and specify the origin bus stop and origin hex grid id as ORIGIN_BS and ORIGIN_HEX respectively;
    • We create a duplicate tibble data.frame to check for duplicate results, and luckily we see an empty tibble
show code 
od_data <- left_join(odbus_filtered , bus_stop_hexgrid_id,
            by = c("ORIGIN_PT_CODE" = "BUS_STOP_N")) %>%
  rename(ORIGIN_BS = ORIGIN_PT_CODE,
         ORIGIN_HEX = grid_id)
duplicate <- od_data %>%
  group_by_all() %>%
  filter(n()>1) %>%
  ungroup()
duplicate
# A tibble: 0 × 4
# ℹ 4 variables: ORIGIN_BS <chr>, DESTINATION_PT_CODE <fct>, TRIPS <dbl>,
#   ORIGIN_HEX <int>
show code 
# od_data[!complete.cases(od_data), ]
# 
# bus_stop_hexgrid_id[] 
# hexagon_sf[hexagon_sf$grid_id %in% c(1, 2),]

# bus_stop_hexgrid_id[bus_stop_hexgrid_id$BUS_STOP_N %in% c(03361),]

# 
# raw_bus_stop_sf[raw_bus_stop_sf$BUS_STOP_N %in% c(03361, 03549, 03579, 59009),]
  • Now we repeat the step, appending the destination hex ID
show code 
od_data <- left_join(od_data , bus_stop_hexgrid_id,
            by = c("DESTINATION_PT_CODE" = "BUS_STOP_N")) %>%
  rename(DESTIN_BS = DESTINATION_PT_CODE,
         DESTIN_HEX = grid_id)
duplicate <- od_data %>%
  group_by_all() %>%
  filter(n()>1) %>%
  ungroup()
duplicate
# A tibble: 0 × 5
# ℹ 5 variables: ORIGIN_BS <chr>, DESTIN_BS <chr>, TRIPS <dbl>,
#   ORIGIN_HEX <int>, DESTIN_HEX <int>
show code 
# od_data[!complete.cases(od_data), ]
  • Once again, we have an empty tibble.
  • We finally create the final version of our flow dataset.
    • Just to be safe, we perform a step to remove duplicates using unique()
    • drop_na() removes bus stops for which we have no info. It turns out there are bus stop numbers outside our bus stop dataset, like 03361, 03549, 03579, 59009;
      • Some of these may be new bus stops, like 03361, Gardens by the Bay Stn Exit 2, and 03549
    • We use group_by() to combine rows with the same $ORIGIN_HEX and $DESTIN_HEX, and aggregate the number of trips with summarise()
show code 
od_data <- od_data %>%
  unique() %>%
  drop_na() %>%
  group_by(ORIGIN_HEX, DESTIN_HEX) %>%
  summarise(SUM_TRIPS = sum(TRIPS))
`summarise()` has grouped output by 'ORIGIN_HEX'. You can override using the
`.groups` argument.
show code 
head(od_data)
# A tibble: 6 × 3
# Groups:   ORIGIN_HEX [1]
  ORIGIN_HEX DESTIN_HEX SUM_TRIPS
       <int>      <int>     <dbl>
1          2          6         3
2          2         13        34
3          2         20       182
4          2         37         1
5          2         50        18
6          2         53         1
  • Now we write the output to RDS
show code 
write_rds(od_data, "data/rds/od_data.rds")
od_data <- read_rds("data/rds/od_data.rds")
  • Remove intrazonal flows
show code 
od_data1 <- od_data[od_data$ORIGIN_HEX!=od_data$DESTIN_HEX,]
  • Create desire lines
show code 
flowLine <- od2line(flow = od_data1, 
                    zones = hexagon_sf,
                    zone_code = "grid_id")
Creating centroids representing desire line start and end points.
show code 
summary(od_data1$SUM_TRIPS)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
    1.0     4.0    17.0   129.6    66.0 73720.0 
tmap_mode("plot")
tmap mode set to plotting
tm_shape(hexagon_sf) +
  tm_polygons() +
flowLine %>%  
  filter(SUM_TRIPS >= 1000) %>%  
tm_shape() +
  tm_lines(lwd = "SUM_TRIPS",
           style = "quantile",
           scale = c(0.1, 1, 3, 5, 7, 10),
           n = 6,
           alpha = 0.3)

1.6.2 Identify Bus Trip Details For Each Hexagon grid_id

  • Here, we again use multiple steps to generate bus trip details for each hexagon grid_id;
    1. We use left_join() to add the grid_id to each row of odbus_filtered, since each row has a unique single bus stop ID (i.e. $BUS_STOP_N);
    2. We use select() to retain only the grid_id and the four peak-trips columns;
    3. We combine group_by() and summarise() to aggregate the trips for each peak by grid_id.
  • Finally, we use head() to preview the grid_trips_df tibble dataframe.
show code 
colnames(odbus_filtered)
[1] "ORIGIN_PT_CODE"      "DESTINATION_PT_CODE" "TRIPS"
show code 
grid_trips_df <- left_join(odbus_filtered, bus_stop_hexgrid_id, 
            by = c("ORIGIN_PT_CODE" = "BUS_STOP_N")) %>%
  select(grid_id, 
         TRIPS)  %>%
  group_by(grid_id) %>%
  summarise(
    TRIPS = sum(TRIPS)
    )
Adding missing grouping variables: `ORIGIN_PT_CODE`
show code 
head(grid_trips_df)
# A tibble: 6 × 2
  grid_id TRIPS
    <int> <dbl>
1       2   390
2       3   114
3       4   291
4       5   241
5       6  1905
6       7   299

1.6.3 Combine Bus Trip Details Back Into hexagon_sf

  • Finally, it’s time to recombine bus trip data back into hexagon_sf;
    • We use left_join on $grid_id to add trip data back into hexagon_sf;
    • We add a failsafe mutate() to replace any “NA” values for the columns.
show code 
hexagon_sf <- left_join(hexagon_sf, grid_trips_df, 
            by = 'grid_id' ) %>%
  mutate(
    TRIPS = ifelse(is.na(TRIPS), 0, TRIPS)
    )

head(hexagon_sf)
Simple feature collection with 6 features and 3 fields
Geometry type: POLYGON
Dimension:     XY
Bounding box:  xmin: 3595.122 ymin: 27889.39 xmax: 4907.622 ymax: 31570
Projected CRS: SVY21 / Singapore TM
  grid_id n_bus_stops TRIPS                   raw_hex_grid
1       1           1     0 POLYGON ((3782.622 27889.39...
2       2           1   390 POLYGON ((4157.622 27889.39...
3       3           1   114 POLYGON ((4532.622 28538.91...
4       4           1   291 POLYGON ((4532.622 30487.47...
5       5           1   241 POLYGON ((4532.622 31136.99...
6       6           1  1905 POLYGON ((4720.122 28214.15...

1.7 Exploratory Data Analysis Of Bus Trips, Across Peak Periods, By Hexagons

  • For ggplot, we need data in long format, so we can use gather() on grid_trips_df from Section 1.6.2 to pivot this;
  • We then pipe this into a geom_boxplot() for an exploratory look:
show code 
# gather(grid_trips_df, key = "Peak", value = "Trips", -grid_id) %>%
#   ggplot( aes(x=grid_id, y='TRIPS', fill=grid_id)) +
#   geom_boxplot() + 
#   ggtitle("Boxplot: Trips over peak periods, 2023-Oct data") +
#     xlab("") + 
#     theme(
#       legend.position="none"
#       ) +
#   coord_flip()
Tip

We also observe that number of trips for Weekday Morning & Weekday Afternoon seems to be larger than Weekend Morning and Weekend Evening peak trips. This is also confirmed by the figure in the next section.

This means that we will have to consider Weekday and Weekend peaks on different scales.

  • This is an exceptionally ugly plot, but it shows an important point: there is some serious right skew in our dataset;
  • Clearly, there are some hexagons with exceptionally high trips compared to the rest of the hexagons But, could this be because some hexagons have up to 11 bus stops, while others have 1 or 2?
  • We do a quick scatter plot based on $n_bus_stops to verify this:
show code 
# hexagon_sf %>% 
#   st_drop_geometry() %>% 
#   pivot_longer(cols = starts_with("WEEK"),
#                names_to = "PEAK", values_to = "TRIPS") %>%
#   ggplot( aes(x=TRIPS, y=n_bus_stops, color=PEAK, shape=PEAK)) +
#   geom_point(size=2) +
#   ggtitle("Scatterplot: Trips over peak periods by number of bus stops per hexagon, 2023-Oct data") +
#     theme(legend.position=c(.85, .15),
#           legend.background = element_rect(fill = "transparent"),
#           legend.key.size = unit(0.5, "cm"),  
#           legend.text = element_text(size = 6),
#           legend.title = element_text(size = 8)
#     )
  • Surprising results from our plot! If we consider those with > 100,000 trips as outliers, most of them come from hexagons with between 4-8 bus stops;
  • There is some correlation between number of bus stops and high numbers of trips, but a stronger factor is peak time; Weekday Morning peak trips, followed by Weekday Afternoon peak trips, contribute to the largest outliers.
Tip

I note that these visualizations only scrape the surface of understanding the data. However, this is not the focus of our study; we do these quick visualizations only to provide better context for our study.